Personalized Recommendations using Knowledge Graphs
In this talk, we will discuss some of the recent research on improving the performance of personalized recommender systems by using knowledge graphs (KG) to uncover the long range preferences of users. We investigate three methods for making KG based recommendations using a general-purpose probabilistic logic system called ProPPR. The simplest of the models uses only the links of the graph. This model is then extended to also use the types of the entities to boost its generalization capabilities. Next, we describe a graph based latent factor model, which combines the strengths of latent factorization with graphs. By comparing to a recently proposed state-of-the-art graph recommendation method on two large datasets, we show that our approaches can give large performance improvements. Additionally, we illustrate that knowledge graphs give maximum advantage when the dataset is sparse, and gradually become redundant as more training data becomes available, and hence are most useful in cold-start settings. We will also discuss how a similar mechanism can be used for certain tasks in the biomedical informatics domain.